Corsini, A. | Search bioRxiv

bioRxiv2026

Neural and behavioural manifold dynamics align across interacting individuals

Coordinating actions with others is fundamental for social behaviour, requiring the nervous system to continuously adapt motor output to a partner's evolving behaviour. Yet the neural population principles supporting such coordination remain poorly understood. To address this, we investigated interpersonal coordination across two dual-EEG studies comprising 44 dyads (88 participants) engaged in either instructed finger movement synchronization or spontaneous face-to-face interaction. Combining kinematics-informed deep contrastive learning with dynamical-systems modelling, we identified low-dimensional neural manifolds. These manifolds aligned geometrically and temporally across interacting partners, mirrored their coordinated behaviour, and uncovered interpersonal alignment not captured by traditional synchrony measures. Importantly, these manifolds exhibited flexible attractor-like organization consistent with a synergistic, dynamical account of motor control, with attractor properties that were co-regulated across partners. Collectively, we propose a novel mechanism in which interpersonal coordination emerges through the intermittent updating of internally organized dynamics by a partner's movements. More broadly, our results establish movement-informed latent-space modeling as a framework for uncovering the low-dimensional population dynamics linking neural activity, neuromuscular control and interpersonal coordination.

neuroscience↗

bioRxiv2026

Evidence of predictive information compression in latent space in humans during speech listening

Speech perception requires transforming acoustic input into neural representations that support linguistic understanding, yet its underlying computational principles remain unclear. Classical efficient coding theories posit optimal compression of sensory input, whereas alternative accounts propose that neural systems preferentially encode information that supports prediction. A key open question is whether such predictive encoding operates on fixed inputs or on flexible internal representations. We instantiated three hypothesis models of speech processing: (i) optimal compression with deep autoencoders, (ii) predictive reconstruction with predictive autoencoders, and (iii) predictive information representation via latent-space prediction using contrastive learning. We compared resulting speech latent representations to electroencephalographic (EEG) activity during speech listening. Representations learned under the predictive information objective best explained neural latents. Crucially, only representations that selectively compressed predictive information predicted behavioral performance, suggesting that neural speech representations are structured to encode predictive information in latent space rather than to maximize compression or input prediction.

neuroscience↗